JFYI,
In i386, /dev/console is the same as /dev/consolectl, and all I/O
operations for /dev/console, /dev/concolectl and /dev/ttyv0 take place
in the screen #0, as shown below. In alpha /dev/console is /dev/ttyv0.
Access to /dev/console is routed to /dev/consolectl's dev_t by
cdevsw functions in kern/tty_cons.c.
When syscons is not acting as the system/kernel console, /dev/console
is not connected to /dev/console or /dev/ttyv0. But /dev/consolectl
still exists.
/dev/console /dev/consolectl /dev/ttyv0 /dev/ttyvN
| | | |
V | | |
dev_t | | |
| V V V
+--------------->dev_t dev_t dev_t
| | |
V V V
struct tty struct tty struct tty
| | |
+--------------->| |
| |
V V
screen #0 screen #N
As /dev/consolectl doesn't need to get input from or put output to the
screen, it can exist without an associated screen. In that sense, it
perhaps doesn't even need a struct tty. Maybe we also should give
/dev/console a separate a screen than ttyv0. See the figure below.
But, even in this configuration, we still have two copies of dev_t for
/dev/console: the native dev_t (*1) and the underlaying dev_t (*2).
As make_dev(9) requires us to supply the name to create a dev_t, dev_t
(*2) will appear under /dev and will be accessible from the user land,
thus, we still have "the last close()" problem, I guess.
/dev/console /dev/consolectl /dev/ttyv0 /dev/ttyvN
| | | |
V | | |
dev_t | | |
| *1 | | |
+-------+ | | |
| | | |
V V V V
dev_t dev_t dev_t dev_t
| *2 | | |
V V V V
struct tty (struct tty) struct tty struct tty
| | |
+-------????--------------->| |
| | |
??? | |
| | |
V V V
screen #X screen #0 screen #N
Kazu
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
